Roleplay Bot chats with local AI characters

This is a neat local-AI demo, but it also shows how fast roleplay bots run into latency and state-management limits once you move beyond a single prompt. The project creates a dedicated roleplay-model from llama3.2, which keeps the character setup self-contained and easy to reproduce. Each turn is sent as a fresh Ollama request with only a one-time persona/scene prefix, so the bot does not persist real multi-turn memory. The cleanup path relies on a Python destructor hook to delete the custom model, which is neat but brittle if the process exits abruptly. The prompt explicitly tells the model to stay in character and ignore restrictive safety guidance, which makes the output vivid but risky to reuse as-is. Source links: Reddit demo https://www.reddit.com/r/LocalLLaMA/comments/1s2j6bl/made-a-roleplaying-chatbot-with-python-and-ollama/; GitHub repo https://github.com/rakin406/roleplay_bot; v1.0.0 release https://github.com/rakin406/roleplay_bot/releases/tag/v1.0.0.